Torus uses spare grid connections to bring modular data centers online in 12–16 weeks

Torus has launched a modular computing system designed to bring inference capacity online at existing commercial and industrial sites, using available grid connections instead of waiting for new data center infrastructure.

The Salt Lake City-based energy and infrastructure company unveiled Foundry on November 11, alongside a $500 million funding announcement led by Magnetar. The funding will support manufacturing expansion and deployments of its energy storage and computing systems across the US.

The launch comes as demand for inference computing grows. Unlike model training, which involves developing and refining models, inference runs those models to generate responses and perform tasks. McKinsey projects that data center demand for inference will grow at a compound annual rate of 35 percent through 2030, compared with 22 percent for training.

Bringing compute to power

Foundry combines computing hardware, power systems, energy storage, cooling and networking in a prefabricated modular unit. Torus says each site can support between 1 and 20 megawatts of computing capacity and can move from contract to operational workloads in 12 to 16 weeks.

The system is designed to operate at existing commercial and industrial sites with available electrical capacity. Rather than waiting for new grid connections or large-scale infrastructure upgrades, Torus aims to use existing interconnections and manage electricity demand through its Torus Reserve energy storage system.

Reserve combines battery storage and inertial-based technology to manage peak demand, provide backup power, and adjust electricity consumption. The company says its load-management controls can help accommodate the fluctuating power requirements of computing hardware while supporting grid stability.

Foundry is also designed for distributed deployment, placing computing resources closer to businesses and the workloads they serve. Torus argues this could appeal to enterprises seeking greater control over where their data is processed, alongside security and intellectual property protections.

Scaling modular capacity

Torus is offering Foundry through three deployment models. The Strategic option provides a powered enclosure without GPUs, while the Partner configuration includes computing hardware. Its Direct model offers dedicated, single-tenant inference capacity for enterprise customers.

The company named AI infrastructure provider Hydra Host among Foundry’s first customers, alongside another unnamed cloud provider. Torus says thousands of GPUs are already deployed in Foundry units and plans to increase that number to more than 47,000 by the end of 2027.

The systems are manufactured at GigaOne, Torus’s 545,000-square-foot facility in Salt Lake City, Utah. The company says its Foundry production line is designed to support 500 megawatts of IT load, with plans to expand capacity to 1 gigawatt.

Torus also says Foundry uses a closed-loop, waterless cooling system that consumes 95 percent less water than evaporative cooling systems. These environmental and performance figures are company-reported claims.

The $500 million financing round was led by Magnetar, with participation from ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital.

The launch reflects a growing effort to address data center power constraints by distributing computing capacity across existing sites. However, Foundry’s deployment timelines, available grid capacity, and planned GPU expansion remain dependent on individual site conditions and future execution.